摘要
以某硫酸法钛白生产线的3608条数据为样本,采用皮尔逊系数和统计P值考察了工业钛液的五个属性变量与偏钛酸粒度D_(50)的相关性,并采用LOF算法对数据进行离异值处理以提高数据质量。在此基础上,采用python语言基于Ridge(岭回归)、Lasso(套索回归)、KNN(K-近邻)、ANN(人工神经网络)、Random Forest(随机森林)、SVR(支持向量机)六种模型编写了偏钛酸粒度控制的回归模型算法,六种算法分别应用在整套数据上的回归预测效果差别不大,离异值处理后数据的RMSE都是在0.276上下波动,MAE则是在0.197上下波动,模型效果均优于离异值处理前模型效果。进一步的,通过对ANN、Random Forest、SVR三个模型进行集成学习模型搭建,回归预测效果得到显著提升,RMSE和MAE值分别降至0.245和0.192。
Taking 3608 data of a titanium dioxide production line by sulfuric acid process as samples,the correlation between the five attribute variables of the industrial titanium sulfate solution and the particle size of metatitanic acid D_(50) was investigated by Pearson coefficient and statistical p value,and the LOF algorithm was used to clean the outlier data and improve the data quality.On this basis,the regression model algorithm of metatitanic acid particle size control was compiled by Python language based on six models of Ridge,Lasso,KNN,ANN,Random forest and SVR.The regression prediction results of the six algorithms applied to the whole set of data have no significant difference.After the outlier processing,the RMSE and MAE of the data fluctuate around 0.276 and 0.197 respectively,and the model effect is better than that of the model before the outlier processing.Furthermore,through the ensemble learning model of ANN,Random Forest and SVR,the regression prediction effect can be significantly improved,and the RMSE and MAE values decreases to 0.245 and 0.192 respectively.
作者
路瑞芳
刘婵
孙伟
吴健春
孙蔷
Lu Ruifang;Liu Chan;Sun Wei;Wu Jianchun;Sun Qiang(State Key Laboratoty of Vanadium and Titanium Resources Comprehensive Utilization,Pangang Group Research tute Co.,Ltd.,Panzhihua 617000,Sichuan,China;School of Chemistry and Chemical Engineering,Chongqing University,Chongqing 400030,China;School of Metallurgy,Northeastern University,Shenyang 110006,Liaoning,China)
出处
《钢铁钒钛》
CAS
北大核心
2021年第2期36-42,共7页
Iron Steel Vanadium Titanium
关键词
钛白
偏钛酸
粒度
软测量
机器学习
TiO_(2)
metatitanic acid
particle size
soft sensing
machine learning